> AI agents: this is one page from PostHog's docs. Full index of Markdown docs for LLMs: https://posthog.com/llms.txt

# Data quality

**Data quality is in alpha**

Data quality isn't available to every project. [Contact support](https://app.posthog.com/home#supportModal) to request access or share feedback.

A query can run successfully and still return the wrong answer. A source can stop updating, a join key can become null, duplicate rows can change a total, or a status column can gain a value that downstream logic doesn't handle. SQL reports none of these as query errors.

Data quality checks turn those assumptions into assertions that PostHog can run repeatedly. You can add checks to:

-   A connected warehouse table
-   A saved view or materialized view
-   A [semantic layer metric](/docs/semantic-layer/metrics.md)
-   One of the PostHog tables `events`, `persons`, or `groups`

For example, checks can assert that:

-   `orders.customer_id` is never null.
-   `orders.order_id` is unique.
-   Every non-null `orders.customer_id` exists in `customers.id`.
-   `orders.status` contains only `paid`, `pending`, or `refunded`.
-   A source has received data within the last 60 minutes.
-   A governed revenue metric never returns a negative value.

Most checks compile to a HogQL query that selects the rows that break the assertion. The check passes when that query returns no rows. PostHog stores the result and failure count, not a snapshot of the failing rows.

## How data quality fits with governance

The [semantic layer](/docs/semantic-layer.md) governs what data means. It records canonical metric definitions, trusted tables, owners, and accepted relationships. Data quality verifies that the data still satisfies the assumptions behind those definitions.

You need both. An approved `monthly_recurring_revenue` metric can use the agreed definition and still return an incorrect number if its source stopped syncing or a join started dropping rows. A passing check also doesn't make a metric canonical or certify a table. It only confirms the assertion that the check ran.

## How checks run

Checks run in several ways:

-   Run one check, every check on a table, view, or metric, or every check in the project manually.
-   Run connected-table checks after that table completes a source sync.
-   Run view checks when PostHog refreshes or materializes the view.
-   Run metric and PostHog-table checks on a recurring schedule.

Only enabled checks run. See [triggers and schedules](/docs/data-warehouse/data-quality/triggers.md) for the exact behavior.

## Severity and health

Set each check's severity based on what a failure means:

-   **Error** – The data shouldn't be trusted. A failure marks the table, view, or metric as failing and sends a notification.
-   **Warning** – The result is worth reviewing, but it doesn't mean consumers must stop using the data.

PostHog rolls the newest results into an overall health status:

-   **Healthy** – At least one enabled check passed, and none failed or errored.
-   **Warning** – Only warning-severity checks are failing.
-   **Failing** – At least one error-severity check is failing.
-   **Erroring** – A check couldn't compile or execute. This doesn't prove that the data is good or bad.
-   **Not run yet** – No enabled check has produced a result.

## Next steps

-   [Create your first check](/docs/data-warehouse/data-quality/start-here.md)
-   [Choose a check type](/docs/data-warehouse/data-quality/check-types.md)
-   [Configure triggers and schedules](/docs/data-warehouse/data-quality/triggers.md)

### Still have questions?

Ask PostHog AI

### Was this page useful?

HelpfulCould be better